Rivindu Perera


RealTextasg: A Model to Present Answers Utilizing the Linguistic Structure of Source Question

Abstract
Recent trends in Question Answering (QA) have led to numerous studies focusing on presenting answers in a form which closely resembles a human generated answer. These studies have used a range of techniques which use the structure of knowledge, generic linguistic structures and template based approaches to construct answers as close as possible to a human generate answer, referred to as human competitive answers. This paper reports the results of an empirical study which uses the linguistic structure of the source question as the basis for a human competitive answer. We propose a typed dependency based approach to generate an answer sentence where linguistic structure of the question is transformed and realized into a sentence containing the answer. We employ the factoid questions from QALD-2 training question set to extract typed dependency patterns based on the root of the parse tree. Using identified patterns we generate a rule set which is used to generate a natural language sentence containing the answer extracted from a knowledge source, realized into a linguistically correct sentence. The evaluation of the approach is performed using QALD-2 testing factoid questions sets with a 78.84% accuracy. The top-10 patterns extracted from training dataset were able to cover 69.19% of test questions.